__poll_t events;
bool done;
bool canceled;
- bool update_events;
- bool update_user_data;
- union {
- struct wait_queue_entry wait;
- struct {
- u64 old_user_data;
- u64 new_user_data;
- };
- };
+ struct wait_queue_entry wait;
};
-struct io_poll_remove {
+struct io_poll_update {
struct file *file;
- u64 addr;
+ u64 old_user_data;
+ u64 new_user_data;
+ __poll_t events;
+ bool update_events;
+ bool update_user_data;
};
struct io_close {
struct file *file;
struct io_rw rw;
struct io_poll_iocb poll;
- struct io_poll_remove poll_remove;
+ struct io_poll_update poll_update;
struct io_accept accept;
struct io_sync sync;
struct io_cancel cancel;
}
}
+static void io_refs_resurrect(struct percpu_ref *ref, struct completion *compl)
+{
+ bool got = percpu_ref_tryget(ref);
+
+ /* already at zero, wait for ->release() */
+ if (!got)
+ wait_for_completion(compl);
+ percpu_ref_resurrect(ref);
+ if (got)
+ percpu_ref_put(ref);
+}
+
static bool io_match_task(struct io_kiocb *head,
struct task_struct *task,
struct files_struct *files)
static inline void req_set_fail_links(struct io_kiocb *req)
{
- if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK)
+ if (req->flags & REQ_F_LINK)
req->flags |= REQ_F_FAIL_LINK;
}
}
static void io_kill_timeout(struct io_kiocb *req, int status)
+ __must_hold(&req->ctx->completion_lock)
{
struct io_timeout_data *io = req->async_data;
- int ret;
- ret = hrtimer_try_to_cancel(&io->timer);
- if (ret != -1) {
+ if (hrtimer_try_to_cancel(&io->timer) != -1) {
atomic_set(&req->ctx->cq_timeouts,
atomic_read(&req->ctx->cq_timeouts) + 1);
list_del_init(&req->timeout.list);
unsigned int cflags)
{
struct io_ring_ctx *ctx = req->ctx;
+ struct io_overflow_cqe *ocqe;
- if (!atomic_read(&req->task->io_uring->in_idle)) {
- struct io_overflow_cqe *ocqe;
-
- ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT);
- if (!ocqe)
- goto overflow;
- if (list_empty(&ctx->cq_overflow_list)) {
- set_bit(0, &ctx->sq_check_overflow);
- set_bit(0, &ctx->cq_check_overflow);
- ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW;
- }
- ocqe->cqe.user_data = req->user_data;
- ocqe->cqe.res = res;
- ocqe->cqe.flags = cflags;
- list_add_tail(&ocqe->list, &ctx->cq_overflow_list);
- return true;
+ ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT);
+ if (!ocqe) {
+ /*
+ * If we're in ring overflow flush mode, or in task cancel mode,
+ * or cannot allocate an overflow entry, then we need to drop it
+ * on the floor.
+ */
+ WRITE_ONCE(ctx->rings->cq_overflow, ++ctx->cached_cq_overflow);
+ return false;
}
-overflow:
- /*
- * If we're in ring overflow flush mode, or in task cancel mode,
- * or cannot allocate an overflow entry, then we need to drop it
- * on the floor.
- */
- WRITE_ONCE(ctx->rings->cq_overflow, ++ctx->cached_cq_overflow);
- return false;
+ if (list_empty(&ctx->cq_overflow_list)) {
+ set_bit(0, &ctx->sq_check_overflow);
+ set_bit(0, &ctx->cq_check_overflow);
+ ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW;
+ }
+ ocqe->cqe.user_data = req->user_data;
+ ocqe->cqe.res = res;
+ ocqe->cqe.flags = cflags;
+ list_add_tail(&ocqe->list, &ctx->cq_overflow_list);
+ return true;
}
static inline bool __io_cqring_fill_event(struct io_kiocb *req, long res,
*/
if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) {
struct io_timeout_data *io = link->async_data;
- int ret;
io_remove_next_linked(req);
link->timeout.head = NULL;
- ret = hrtimer_try_to_cancel(&io->timer);
- if (ret != -1) {
+ if (hrtimer_try_to_cancel(&io->timer) != -1) {
io_cqring_fill_event(link, -ECANCELED, 0);
io_put_req_deferred(link, 1);
return true;
if (likely(req->flags & REQ_F_LINK_TIMEOUT))
posted = io_kill_linked_timeout(req);
- if (unlikely(req->flags & REQ_F_FAIL_LINK)) {
+ if (unlikely((req->flags & REQ_F_FAIL_LINK) &&
+ !(req->flags & REQ_F_HARDLINK))) {
posted |= (req->link != NULL);
io_fail_links(req);
}
return ret;
}
-/*
- * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a
- * non-spinning poll check - we'll still enter the driver poll loop, but only
- * as a non-spinning completion check.
- */
-static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events,
- long min)
-{
- while (!list_empty(&ctx->iopoll_list) && !need_resched()) {
- int ret;
-
- ret = io_do_iopoll(ctx, nr_events, min);
- if (ret < 0)
- return ret;
- if (*nr_events >= min)
- return 0;
- }
-
- return 1;
-}
-
/*
* We can't just wait for polled events to come to us, we have to actively
* find and complete them.
static int io_iopoll_check(struct io_ring_ctx *ctx, long min)
{
unsigned int nr_events = 0;
- int iters = 0, ret = 0;
+ int ret = 0;
/*
* We disallow the app entering submit/complete with polling, but we
* that got punted to a workqueue.
*/
mutex_lock(&ctx->uring_lock);
+ /*
+ * Don't enter poll loop if we already have events pending.
+ * If we do, we can potentially be spinning for commands that
+ * already triggered a CQE (eg in error).
+ */
+ if (test_bit(0, &ctx->cq_check_overflow))
+ __io_cqring_overflow_flush(ctx, false);
+ if (io_cqring_events(ctx))
+ goto out;
do {
- /*
- * Don't enter poll loop if we already have events pending.
- * If we do, we can potentially be spinning for commands that
- * already triggered a CQE (eg in error).
- */
- if (test_bit(0, &ctx->cq_check_overflow))
- __io_cqring_overflow_flush(ctx, false);
- if (io_cqring_events(ctx))
- break;
-
/*
* If a submit got punted to a workqueue, we can have the
* application entering polling for a command before it gets
* forever, while the workqueue is stuck trying to acquire the
* very same mutex.
*/
- if (!(++iters & 7)) {
+ if (list_empty(&ctx->iopoll_list)) {
mutex_unlock(&ctx->uring_lock);
io_run_task_work();
mutex_lock(&ctx->uring_lock);
- }
-
- ret = io_iopoll_getevents(ctx, &nr_events, min);
- if (ret <= 0)
- break;
- ret = 0;
- } while (min && !nr_events && !need_resched());
+ if (list_empty(&ctx->iopoll_list))
+ break;
+ }
+ ret = io_do_iopoll(ctx, &nr_events, min);
+ } while (!ret && nr_events < min && !need_resched());
+out:
mutex_unlock(&ctx->uring_lock);
return ret;
}
/*
* After the iocb has been issued, it's safe to be found on the poll list.
* Adding the kiocb to the list AFTER submission ensures that we don't
- * find it from a io_iopoll_getevents() thread before the issuer is done
+ * find it from a io_do_iopoll() thread before the issuer is done
* accessing the kiocb cookie.
*/
static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async)
poll->head = NULL;
poll->done = false;
poll->canceled = false;
- poll->update_events = poll->update_user_data = false;
#define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP)
/* mask in events that we always want/need */
poll->events = events | IO_POLL_UNMASK;
pt->error = -EINVAL;
return;
}
+ /*
+ * Can't handle multishot for double wait for now, turn it
+ * into one-shot mode.
+ */
+ if (!(req->poll.events & EPOLLONESHOT))
+ req->poll.events |= EPOLLONESHOT;
/* double add on the same waitqueue head, ignore */
if (poll->head == head)
return;
bool do_complete;
io_poll_remove_double(req);
+ do_complete = __io_poll_remove_one(req, io_poll_get_single(req), true);
- if (req->opcode == IORING_OP_POLL_ADD) {
- do_complete = __io_poll_remove_one(req, &req->poll, true);
- } else {
+ if (req->opcode != IORING_OP_POLL_ADD && do_complete) {
struct async_poll *apoll = req->apoll;
/* non-poll requests have submit ref still */
- do_complete = __io_poll_remove_one(req, &apoll->poll, true);
- if (do_complete) {
- req_ref_put(req);
- kfree(apoll->double_poll);
- kfree(apoll);
- }
+ req_ref_put(req);
+ kfree(apoll->double_poll);
+ kfree(apoll);
}
-
return do_complete;
}
return posted != 0;
}
-static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr)
+static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr,
+ bool poll_only)
__must_hold(&ctx->completion_lock)
{
struct hlist_head *list;
hlist_for_each_entry(req, list, hash_node) {
if (sqe_addr != req->user_data)
continue;
+ if (poll_only && req->opcode != IORING_OP_POLL_ADD)
+ continue;
return req;
}
-
return NULL;
}
-static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr)
+static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr,
+ bool poll_only)
__must_hold(&ctx->completion_lock)
{
struct io_kiocb *req;
- req = io_poll_find(ctx, sqe_addr);
+ req = io_poll_find(ctx, sqe_addr, poll_only);
if (!req)
return -ENOENT;
if (io_poll_remove_one(req))
return -EALREADY;
}
-static int io_poll_remove_prep(struct io_kiocb *req,
+static __poll_t io_poll_parse_events(const struct io_uring_sqe *sqe,
+ unsigned int flags)
+{
+ u32 events;
+
+ events = READ_ONCE(sqe->poll32_events);
+#ifdef __BIG_ENDIAN
+ events = swahw32(events);
+#endif
+ if (!(flags & IORING_POLL_ADD_MULTI))
+ events |= EPOLLONESHOT;
+ return demangle_poll(events) | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT));
+}
+
+static int io_poll_update_prep(struct io_kiocb *req,
const struct io_uring_sqe *sqe)
{
+ struct io_poll_update *upd = &req->poll_update;
+ u32 flags;
+
if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
return -EINVAL;
- if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index ||
- sqe->poll_events)
+ if (sqe->ioprio || sqe->buf_index)
+ return -EINVAL;
+ flags = READ_ONCE(sqe->len);
+ if (flags & ~(IORING_POLL_UPDATE_EVENTS | IORING_POLL_UPDATE_USER_DATA |
+ IORING_POLL_ADD_MULTI))
+ return -EINVAL;
+ /* meaningless without update */
+ if (flags == IORING_POLL_ADD_MULTI)
return -EINVAL;
- req->poll_remove.addr = READ_ONCE(sqe->addr);
- return 0;
-}
-
-/*
- * Find a running poll command that matches one specified in sqe->addr,
- * and remove it if found.
- */
-static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags)
-{
- struct io_ring_ctx *ctx = req->ctx;
- int ret;
+ upd->old_user_data = READ_ONCE(sqe->addr);
+ upd->update_events = flags & IORING_POLL_UPDATE_EVENTS;
+ upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA;
- spin_lock_irq(&ctx->completion_lock);
- ret = io_poll_cancel(ctx, req->poll_remove.addr);
- spin_unlock_irq(&ctx->completion_lock);
+ upd->new_user_data = READ_ONCE(sqe->off);
+ if (!upd->update_user_data && upd->new_user_data)
+ return -EINVAL;
+ if (upd->update_events)
+ upd->events = io_poll_parse_events(sqe, flags);
+ else if (sqe->poll32_events)
+ return -EINVAL;
- if (ret < 0)
- req_set_fail_links(req);
- __io_req_complete(req, issue_flags, ret, 0);
return 0;
}
static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
{
struct io_poll_iocb *poll = &req->poll;
- u32 events, flags;
+ u32 flags;
if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
return -EINVAL;
- if (sqe->ioprio || sqe->buf_index)
+ if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->addr)
return -EINVAL;
flags = READ_ONCE(sqe->len);
- if (flags & ~(IORING_POLL_ADD_MULTI | IORING_POLL_UPDATE_EVENTS |
- IORING_POLL_UPDATE_USER_DATA))
- return -EINVAL;
- events = READ_ONCE(sqe->poll32_events);
-#ifdef __BIG_ENDIAN
- events = swahw32(events);
-#endif
- if (!(flags & IORING_POLL_ADD_MULTI))
- events |= EPOLLONESHOT;
- poll->update_events = poll->update_user_data = false;
- if (flags & IORING_POLL_UPDATE_EVENTS) {
- poll->update_events = true;
- poll->old_user_data = READ_ONCE(sqe->addr);
- }
- if (flags & IORING_POLL_UPDATE_USER_DATA) {
- poll->update_user_data = true;
- poll->new_user_data = READ_ONCE(sqe->off);
- }
- if (!(poll->update_events || poll->update_user_data) &&
- (sqe->off || sqe->addr))
+ if (flags & ~IORING_POLL_ADD_MULTI)
return -EINVAL;
- poll->events = demangle_poll(events) |
- (events & (EPOLLEXCLUSIVE|EPOLLONESHOT));
+
+ poll->events = io_poll_parse_events(sqe, flags);
return 0;
}
-static int __io_poll_add(struct io_kiocb *req)
+static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags)
{
struct io_poll_iocb *poll = &req->poll;
struct io_ring_ctx *ctx = req->ctx;
return ipt.error;
}
-static int io_poll_update(struct io_kiocb *req)
+static int io_poll_update(struct io_kiocb *req, unsigned int issue_flags)
{
struct io_ring_ctx *ctx = req->ctx;
struct io_kiocb *preq;
int ret;
spin_lock_irq(&ctx->completion_lock);
- preq = io_poll_find(ctx, req->poll.old_user_data);
+ preq = io_poll_find(ctx, req->poll_update.old_user_data, true);
if (!preq) {
ret = -ENOENT;
goto err;
- } else if (preq->opcode != IORING_OP_POLL_ADD) {
- /* don't allow internal poll updates */
- ret = -EACCES;
+ }
+
+ if (!req->poll_update.update_events && !req->poll_update.update_user_data) {
+ completing = true;
+ ret = io_poll_remove_one(preq) ? 0 : -EALREADY;
goto err;
}
return 0;
}
/* only mask one event flags, keep behavior flags */
- if (req->poll.update_events) {
+ if (req->poll_update.update_events) {
preq->poll.events &= ~0xffff;
- preq->poll.events |= req->poll.events & 0xffff;
+ preq->poll.events |= req->poll_update.events & 0xffff;
preq->poll.events |= IO_POLL_UNMASK;
}
- if (req->poll.update_user_data)
- preq->user_data = req->poll.new_user_data;
-
+ if (req->poll_update.update_user_data)
+ preq->user_data = req->poll_update.new_user_data;
spin_unlock_irq(&ctx->completion_lock);
/* complete update request, we're done with it */
io_req_complete(req, ret);
if (!completing) {
- ret = __io_poll_add(preq);
+ ret = io_poll_add(preq, issue_flags);
if (ret < 0) {
req_set_fail_links(preq);
io_req_complete(preq, ret);
return 0;
}
-static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags)
-{
- if (!req->poll.update_events && !req->poll.update_user_data)
- return __io_poll_add(req);
- return io_poll_update(req);
-}
-
static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
{
struct io_timeout_data *data = container_of(timer,
{
struct io_timeout_data *io;
struct io_kiocb *req;
- int ret = -ENOENT;
+ bool found = false;
list_for_each_entry(req, &ctx->timeout_list, timeout.list) {
- if (user_data == req->user_data) {
- ret = 0;
+ found = user_data == req->user_data;
+ if (found)
break;
- }
}
-
- if (ret == -ENOENT)
- return ERR_PTR(ret);
+ if (!found)
+ return ERR_PTR(-ENOENT);
io = req->async_data;
- ret = hrtimer_try_to_cancel(&io->timer);
- if (ret == -1)
+ if (hrtimer_try_to_cancel(&io->timer) == -1)
return ERR_PTR(-EALREADY);
list_del_init(&req->timeout.list);
return req;
ret = io_timeout_cancel(ctx, sqe_addr);
if (ret != -ENOENT)
goto done;
- ret = io_poll_cancel(ctx, sqe_addr);
+ ret = io_poll_cancel(ctx, sqe_addr, false);
done:
if (!ret)
ret = success_ret;
ret = io_timeout_cancel(ctx, sqe_addr);
if (ret != -ENOENT)
goto done;
- ret = io_poll_cancel(ctx, sqe_addr);
+ ret = io_poll_cancel(ctx, sqe_addr, false);
if (ret != -ENOENT)
goto done;
spin_unlock_irq(&ctx->completion_lock);
case IORING_OP_POLL_ADD:
return io_poll_add_prep(req, sqe);
case IORING_OP_POLL_REMOVE:
- return io_poll_remove_prep(req, sqe);
+ return io_poll_update_prep(req, sqe);
case IORING_OP_FSYNC:
return io_fsync_prep(req, sqe);
case IORING_OP_SYNC_FILE_RANGE:
ret = io_poll_add(req, issue_flags);
break;
case IORING_OP_POLL_REMOVE:
- ret = io_poll_remove(req, issue_flags);
+ ret = io_poll_update(req, issue_flags);
break;
case IORING_OP_SYNC_FILE_RANGE:
ret = io_sync_file_range(req, issue_flags);
fput(file);
}
#endif
+ io_free_file_tables(&ctx->file_table, ctx->nr_user_files);
+ kfree(ctx->file_data);
+ ctx->file_data = NULL;
+ ctx->nr_user_files = 0;
}
static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx)
static int io_sqe_files_unregister(struct io_ring_ctx *ctx)
{
- struct io_rsrc_data *data = ctx->file_data;
int ret;
- if (!data)
+ if (!ctx->file_data)
return -ENXIO;
- ret = io_rsrc_ref_quiesce(data, ctx);
- if (ret)
- return ret;
-
- __io_sqe_files_unregister(ctx);
- io_free_file_tables(&ctx->file_table, ctx->nr_user_files);
- kfree(data);
- ctx->file_data = NULL;
- ctx->nr_user_files = 0;
- return 0;
+ ret = io_rsrc_ref_quiesce(ctx->file_data, ctx);
+ if (!ret)
+ __io_sqe_files_unregister(ctx);
+ return ret;
}
static void io_sq_thread_unpark(struct io_sq_data *sqd)
static void io_sq_thread_stop(struct io_sq_data *sqd)
{
WARN_ON_ONCE(sqd->thread == current);
+ WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state));
- mutex_lock(&sqd->lock);
set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
+ mutex_lock(&sqd->lock);
if (sqd->thread)
wake_up_process(sqd->thread);
mutex_unlock(&sqd->lock);
ret = io_sqe_files_scm(ctx);
if (ret) {
- io_sqe_files_unregister(ctx);
+ __io_sqe_files_unregister(ctx);
return ret;
}
}
mutex_lock(&ctx->uring_lock);
- io_sqe_files_unregister(ctx);
+ if (ctx->file_data) {
+ if (!atomic_dec_and_test(&ctx->file_data->refs))
+ wait_for_completion(&ctx->file_data->done);
+ __io_sqe_files_unregister(ctx);
+ }
if (ctx->rings)
__io_cqring_overflow_flush(ctx, true);
mutex_unlock(&ctx->uring_lock);
WARN_ON_ONCE(time_after(jiffies, timeout));
} while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20));
+ init_completion(&exit.completion);
+ init_task_work(&exit.task_work, io_tctx_exit_cb);
+ exit.ctx = ctx;
/*
* Some may use context even when all refs and requests have been put,
* and they are free to do so while still holding uring_lock or
node = list_first_entry(&ctx->tctx_list, struct io_tctx_node,
ctx_node);
- exit.ctx = ctx;
- init_completion(&exit.completion);
- init_task_work(&exit.task_work, io_tctx_exit_cb);
+ /* don't spin on a single task if cancellation failed */
+ list_rotate_left(&ctx->tctx_list);
ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL);
if (WARN_ON_ONCE(ret))
continue;
mutex_unlock(&ctx->uring_lock);
wait_for_completion(&exit.completion);
- cond_resched();
mutex_lock(&ctx->uring_lock);
}
mutex_unlock(&ctx->uring_lock);
if (ret < 0)
break;
} while (1);
-
mutex_lock(&ctx->uring_lock);
if (ret) {
- percpu_ref_resurrect(&ctx->refs);
- goto out_quiesce;
+ io_refs_resurrect(&ctx->refs, &ctx->ref_comp);
+ return ret;
}
}
if (io_register_op_must_quiesce(opcode)) {
/* bring the ctx back to life */
percpu_ref_reinit(&ctx->refs);
-out_quiesce:
reinit_completion(&ctx->ref_comp);
}
return ret;